Cadence Maestro: chip design models customers can post-train and run on-prem

Cadence, one of the two big chip design software (EDA) vendors, has announced Maestro, a set of AI models built for its chip design agents. It is a portfolio rather than one model: post-trained domain LLMs, trained decision models, and agent skills that Cadence says improve themselves. They power its AuraStack, ChipStack, InnoStack and ViraStack agents.

How it is built

Maestro is built partly on Nvidia’s open-weight Nemotron models and uses Nvidia OpenShell for agent security. Training uses reinforcement learning against Cadence’s own deterministic solvers, which check outputs against engineering goals, constraints and ground truth, and feed the results back into the models. That is the core idea: chip tools already know how to score a design, so they can grade the model.

Cadence’s initial figures are up to 30% better quality of results and up to 5x token efficiency compared with a general-purpose foundation model. It has not published the benchmark behind them, and Tom’s Hardware notes that “quality of results” in chip design usually means power, performance and area.

Where your design data goes

Customers can run Maestro on Cadence’s cloud, a public cloud, or on-premises, and post-train it on their own designs and workflows. Paul Cunningham, who runs the business, told Tom’s Hardware that customers could run that fine-tuning with Cadence’s harness “on their own GPUs, on their own network.” The Next Platform describes customization as offered more as a managed service, so the exact setup may vary by customer.

That is the contrast Reuters draws with rival Synopsys, whose GPT-Synopsys model, announced with OpenAI on 30 September, runs on OpenAI-hosted infrastructure. For chip companies, design files are among their most guarded assets.

Availability and price

The first models go to select early-access customers in Q4 2026. Cunningham told Tom’s Hardware that Maestro use counts toward Cadence’s agent pricing, a subscription with a usage allowance plus overage charges; he did not say whether the token savings lower the bill. Socionext says it is putting InnoStack into production design work.